Inst4DGS: Instance-Decomposed 4D Gaussian Splatting with Multi-Video Label Permutation Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lee, Yonghan, Manocha, Dinesh
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915874228666368
author Lee, Yonghan
Manocha, Dinesh
author_facet Lee, Yonghan
Manocha, Dinesh
contents We present Inst4DGS, an instance-decomposed 4D Gaussian Splatting (4DGS) approach with long-horizon per-Gaussian trajectories. While dynamic 4DGS has advanced rapidly, instance-decomposed 4DGS remains underexplored, largely due to the difficulty of associating inconsistent instance labels across independently segmented multi-view videos. We address this challenge by introducing per-video label-permutation latents that learn cross-video instance matches through a differentiable Sinkhorn layer, enabling direct multi-view supervision with consistent identity preservation. This explicit label alignment yields sharp decision boundaries and temporally stable identities without identity drift. To further improve efficiency, we propose instance-decomposed motion scaffolds that provide low-dimensional motion bases per object for long-horizon trajectory optimization. Experiments on Panoptic Studio and Neural3DV show that Inst4DGS jointly supports tracking and instance decomposition while achieving state-of-the-art rendering and segmentation quality. On the Panoptic Studio dataset, Inst4DGS improves PSNR from 26.10 to 28.36, and instance mIoU from 0.6310 to 0.9129, over the strongest baseline.
format Preprint
id arxiv_https___arxiv_org_abs_2603_18402
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Inst4DGS: Instance-Decomposed 4D Gaussian Splatting with Multi-Video Label Permutation Learning
Lee, Yonghan
Manocha, Dinesh
Computer Vision and Pattern Recognition
We present Inst4DGS, an instance-decomposed 4D Gaussian Splatting (4DGS) approach with long-horizon per-Gaussian trajectories. While dynamic 4DGS has advanced rapidly, instance-decomposed 4DGS remains underexplored, largely due to the difficulty of associating inconsistent instance labels across independently segmented multi-view videos. We address this challenge by introducing per-video label-permutation latents that learn cross-video instance matches through a differentiable Sinkhorn layer, enabling direct multi-view supervision with consistent identity preservation. This explicit label alignment yields sharp decision boundaries and temporally stable identities without identity drift. To further improve efficiency, we propose instance-decomposed motion scaffolds that provide low-dimensional motion bases per object for long-horizon trajectory optimization. Experiments on Panoptic Studio and Neural3DV show that Inst4DGS jointly supports tracking and instance decomposition while achieving state-of-the-art rendering and segmentation quality. On the Panoptic Studio dataset, Inst4DGS improves PSNR from 26.10 to 28.36, and instance mIoU from 0.6310 to 0.9129, over the strongest baseline.
title Inst4DGS: Instance-Decomposed 4D Gaussian Splatting with Multi-Video Label Permutation Learning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.18402